Covalency does not suppress O2 formation in 4d and 5d Li-rich O-redox cathodes

Layered Li-rich transition metal oxides undergo O-redox, involving the oxidation of the O2− ions charge compensated by extraction of Li+ ions. Recent results have shown that for 3d transition metal oxides the oxidized O2− forms molecular O2 trapped in the bulk particles. Other forms of oxidised O2− such as O22− or (O–O)n− with long bonds have been proposed, based especially on work on 4 and 5d transition metal oxides, where TM–O bonding is more covalent. Here, we show, using high resolution RIXS that molecular O2 is formed in the bulk particles on O2‒ oxidation in the archetypal Li-rich ruthenates and iridate compounds, Li2RuO3, Li2Ru0.5Sn0.5O3 and Li2Ir0.5Sn0.5O3. The results indicate that O-redox occurs across 3, 4, and 5d transition metal oxides, forming O2, i.e. the greater covalency of the 4d and 5d compounds still favours O2. RIXS and XAS data for Li2IrO3 are consistent with a charge compensation mechanism associated primarily with Ir redox up to and beyond the 5+ oxidation state, with no evidence of O–O dimerization.

L ayered Li-rich 3d transition metal oxide intercalation compounds, such as Li [Li 0.2 Ni 0.13 Co 0.13 Mn 0.54 ]O 2 , have received a great deal of attention because Li + can be extracted beyond the limit of transition metal (TM) oxidation, with the charge being compensated by oxidation of the O 2ions (O-redox) [1][2][3][4][5][6][7][8][9] . These compounds typically possess honeycomb ordered Li and TM within the TM layer. However, Mn-based honeycomb ordered structures do not provide a stable framework for oxidised O 2− and have been shown to undergo extensive TM migration, and bulk O-O dimerization leading to voltage hysteresis and loss of energy density, in addition to surface O 2 evolution [10][11][12][13][14] . It has been shown recently that the dimerised O-O is molecular O 2 , which is trapped in voids within the bulk of the charged particles 15,16 . Molecular O 2 formation is responsible for both surface O-loss and bulk O-redox.
Pioneering work by Tarascon, Doublet and co-workers [17][18][19][20][21][22][23][24][25] and by others [26][27][28][29][30][31] , on the 4d-and 5d-based analogues of the 3d compounds, such as Li 2 RuO 3 , Na 2 RuO 3 , Li 2 Ru 0.5 Mn 0.5 O 3 , Li 2 IrO 3 and Li 2 Ir 0.75 Sn 0. 25 O 3 , has led to important advances in the understanding of O-redox. These systems possess the same honeycomb ordered Li and TM ions within the TM layer. With the exception of Li 2 IrO 3 , they exhibit voltage hysteresis, with a plateau on charge and a low voltage S-shaped profile on subsequent discharge. Loss of honeycomb ordering due to Li/TM disordering accompanies the voltage hysteresis along with O 2 loss from the surface of the particles 17,19,21,26,29 . It has been reported that peroxides O 2 2− and longer O-O dimers form beyond the limits of transition metal oxidation in the 4 and 5d transition metal oxides 17,19 . The more strongly hybridised TM-O bonding of the 4 and 5d transition metals compared with the 3d counterparts has been cited as a reason for stabilising such O-O species with longer O-O bonds [18][19][20]32 . In highly covalent transition metal sulphides, selenides and tellurides, electron holes can be stabilised through dimerization of the chalcogen (S 2 ) 2− which remains coordinated to the transition metal due to the strong orbital overlap 33 .
It has proved very challenging to identify experimentally the form of oxidised O 2− in charged materials. The recent application of high resolution RIXS spectroscopy has proved useful in probing the nature of oxidised O 2−15,16 . Here we apply this technique to the 4 and 5d materials, providing direct evidence for the presence of molecular O 2 , trapped in the bulk of the archetypal 4d and 5d systems Li 2 26 . These data indicate that the more covalent TM-O bonding in 4 and 5d compared with 3d TM oxides still favours the formation of molecular O 2 , helping to explain why O-loss is also observed from the surface of these compounds. The implication is that the O-redox process, involving molecular O 2 formation at the surface and in the bulk, is the same for Li-rich systems with the honeycomb superstructure moving down the Periodic Table.

Results
The Li-rich ruthenates and iridates. Li 2 RuO 3 , Li 2 IrO 3 and Snsubstituted Li 2 Ru 0.5 Sn 0.5 O 3 and Li 2 Ir 0.5 Sn 0.5 O 3 were prepared following the methods of previous reports 17,19 . Powder X-ray Diffraction data, Supplementary Figs. 1-3, confirm the formation of the compounds. Each of the materials possess O3-type layered structures with honeycomb ordering in the TM plane, Fig. 1a manifesting as the familiar superlattice peaks between 2θ = 18°a nd 34°. As seen before, there is evidence of some stacking faults between the ordered layers, which result in asymmetric peak broadening of these superlattice peaks, especially for the Snsubstituted samples.
The first cycle load curves are shown in Fig. 1b, c and are plotted against the nominal oxidation state of Ru and Ir in each case, since Sn 4+ is known to be redox inactive. The electrochemical behaviour is reversible with very little hysteresis when cycling below Ru 5+ and Ir 5+ . When sufficient Li is extracted to exceed the +5 oxidation state on Ru, an extended high voltage plateau is observed followed by an S-shaped discharge. For the iridates, reversibility can be maintained up to +5.5 supported by Ir redox as shown recently by Hong et al. 26 . Further Li can be extracted beyond this limit in Li 2 Ir 0.5 Sn 0.5 O 3 subsequently inducing voltage hysteresis. Li 2 IrO 3 is the only material where TM migration and loss of the honeycomb ordering is avoided upon charging to 4.6 V, in accord with its reversible electrochemical behaviour 26,30 .
Spectroscopic characterisation of O. Understanding O-redox has proven to be a challenge due in part to the need for techniques capable of determining the nature of O species formed in the bulk. In this study, we have employed X-ray absorption spectroscopy (XAS) in partial fluorescence yield (PFY) mode and high resolution resonant inelastic X-ray scattering (RIXS) at the O Kedge, as they offer a direct probe of the electronic states on O at depths of up to 50-100 nm into the sample. XAS probes the empty states above the Fermi level. In RIXS, excitation of core electrons to empty states above the Fermi level results in emission as electrons from filled valance states below the Fermi level relax to the core-hole states. RIXS complements XAS as it provides a direct probe of the valence states on O.
In Fig. 2, we present the O K-edge XAS and RIXS for Li 2 RuO 3 and Li 2 Ru 0.5 Sn 0.5 O 3 collected ex situ at different points along the load curve on charge and discharge. Considering first the XAS spectra. For Li 2 RuO 3 , on initial charge to the beginning of the plateau, there is a pronounced increase in intensity at the leading edge of the pre-edge (lowest energy peak between 529 and 530 eV) indicating the formation of electron hole states in hybridised Ru-O orbitals consistent with Ru oxidation from +4 to +5, as previously reported 17,25 . Across the plateau there is no further increase in this region but instead new states appear at 531 eV. After discharge, both of these changes are reversed, and the pre-  ARTICLE edge reduces in intensity. The pre-edge for the discharged sample is comparatively broad when compared directly with that of the pristine indicating a rehybridization of the Ru-O bonding between the two samples. The structure has been shown to undergo TM migration during the first cycle and the XAS peak broadening we observe here is consistent with an increase in the local disorder around O. For Li 2 Ru 0.5 Sn 0.5 O 3 , the pre-edge broadening after the plateau on charge and in the discharged material is more pronounced than Li 2 RuO 3 in line with the greater degree of O-redox and more extensive TM migration for the Sn-substituted material 17 .
To interrogate the electronic states formed at 531 eV further, RIXS measurements were performed for each sample at this excitation energy. The emission spectra are plotted as is convention, as energy loss (difference between excitation and emission energy). At the top of charge two new energy loss features become evident, a broad peak at 8 eV and a progression of sharp peaks between 0 and 2 eV, as we observed previously for Li 1 The progression of peaks in the 0-2 eV region correspond to the vibrations of a molecular O 2 diatomic, also shown in Fig. 2 35 . The emission spectra for a range of excitation energies across the O K-edge for the fully charged electrodes were also measured and are presented as RIXS maps. The data show no evidence of any other vibrational features at different excitation energies. After discharge, these new features are much diminished in intensity indicating reversible electrochemical reduction of molecular O 2 has occurred.
The same measurements were also performed for the iridate samples and are presented in Fig. 3. Li 2 Ir 0.5 Sn 0.5 O 3 exhibits very similar changes to those described for Li 2 RuO 3 and Li 2 Ru 0.5 Sn 0.5 O 3 consistent with the voltage profile observed. On the other hand, the RIXS spectra for Li 2 IrO 3 do not show any evidence for the presence of molecular O 2 in the fully charged electrodes. Instead, a strong increase in intensity at the leading edge of the pre-edge is seen when charging Ir beyond the +5 oxidation state. This observation supports the conclusion that the high voltage plateau in Li 2 IrO 3 is associated with Ir rather than O oxidation 26   spectra for the pristine and discharged electrodes are almost fully superimposable indicating minimal irreversible change to the electronic structure and thus structural stability. Since the measurements are performed under ultra-high vacuum (UHV) conditions and the samples had been pumped down overnight under UHV, the electrodes will be fully outgassed, so any molecular species that are detected are trapped within the bulk of the primary particles. To rule out the possible influence of beam damage inducing molecular O 2 , we performed all our measurements at low temperature, 20 K, and conducted measurements at the same sample location over a range of timescales. The data presented in Supplementary Fig. 4 show no change in the peak spacing between spectra acquired after 30 s and 1800 s exposure times and only a minor decrease in intensity, which is in line with our previous beam sensitivity studies for molecular O 2 in Li 1.2 Ni 0.13 Co 0.13 Mn 0.54 O 2 15 . In that paper, we also extensively examined the effect of temperature and photon flux and showed that neither of these factors have a detectable influence on the vibrational peak spacing, reinforcing that the O 2 observed by RIXS is intrinsic to the cathode.
Although the spectroscopic data show no evidence of O species other than O 2 , RIXS data were also collected for KO 2 and Li 2 O 2 to rule out the possibility of O 2− and O 2 2− . The results are presented in Fig. 4 where they are compared directly with the spectra for the Ru and Ir compounds. The peak spacing for Li 2 O 2 around the elastic peak corresponding to the vibrational spectrum is almost exactly half of the peak spacing observed for molecular O 2 in the cathodes as clearly seen in the Birge-Sponer plot, Fig. 4b. This is closely in line with the vibrational frequency for O 2 2− which is well known to be half that of molecular O 2 . For KO 2 , containing the superoxide moiety O 2 − of intermediate bond order to O 2 and O 2 2− , the peak spacing lies halfway between the two. While some differences in the RIXS spectra for different peroxide and superoxide compounds is possible, the vibrational spectra associated with the elastic peak is determined primarily by the O-O bond length/strength and therefore is characteristic of these species in general. The clear distinction that can be made between O 2 , O 2 − and O 2 2− dimers demonstrates the power of high resolution RIXS and provides evidence for the formation of molecular O 2 in the Li-rich cathodes.

Discussion
Studies using X-ray photoelectron spectroscopy (XPS), electron paramagnetic resonance (EPR) and density functional theory (DFT) pointed to the possibility of peroxo-like O 2 n− species, where n = 1, 2 or 3 17,20,25,36 . Scanning transmission spectroscopy (STEM) and neutron powder diffraction studies of Li 2 IrO 3 , suggested longer peroxo-like dimers (3 < n < 3.3) 19 . However, excellent though these studies are, it is a challenge for these techniques to identify unambiguously the nature of the oxidised oxygen species. XPS, being an electron emission technique, is, in general, more limited in its ability to measure bulk species than RIXS, which utilises photon emissions, and XPS can often be strongly influenced by surface contributions. Turning to local structural probes, imaging individual O-O defects in the bulk is beyond the capabilities of current STEM techniques and resolving O-O species at such low interatomic separations and concentrations is very challenging for total scattering data. Since they are magnetically complex materials, the 4d and 5d systems defy clear characterisation of oxidised O by either 17 O NMR or SQUID. In contrast, the high resolution RIXS that we employ in this study is element specific, probes 50-100 nm deep into the particles, and has allowed us to clearly identify molecular O 2 in the bulk of solid materials. High resolution RIXS has already provided evidence that O 2− oxidation in 3d TM oxides forms molecular O 2 trapped in voids in the bulk particles. This observation is further supported by 17 O NMR which not only identifies trapped molecular O 2 as the O-O species formed on charge, but also shows that it is present in quantities commensurate with that expected from the charged passed in Li 1.2 Ni 0.13 Co 0.13 Mn 0.54 O 2 15 .
Our RIXS and XAS results indicate that the O-redox process can be described as molecular O 2 formation throughout the cathode, both as evolved O 2 at the surface, which has already been demonstrated with operando mass spec 19,29 , and trapped O 2 within the bulk. The 4 and 5d transition metal oxides generally exhibit greater covalency in the TM-O bond than those of their 3d counterparts; associated with the greater TMd-O2p overlap and lower electron repulsion of the larger 4 and 5d orbitals. The results presented here indicate that these more covalent systems bear closer resemblance to the 3d Li-rich materials than previously thought and that any greater covalency in the TM-O  However, for Li 2 Ir 0.5 Sn 0.5 O 3 the high resolution RIXS shows the presence of molecular O 2 , rather than peroxide, as the only form of oxidised oxygen species. High resolution RIXS also reveals molecular O 2 is present in the ruthenates in contrast to previous reports of peroxides. The ability of RIXS to show an absence of signal for materials supported exclusively by TM-redox and identify oxidised O when it is present in O-redox materials demonstrates its utility for probing oxidised O species. Ir 5+ , low spin t 2g 4 , has 1 more electron than Ru 5+ , t 2g 3 , and it is spin-paired, Fig. 5a. Removal of this higher energy spin down electron occurs at a lower voltage than for an electron on Ru 5+ . The oxidation of O 2− sits between the energies for Ir 5+/6+ and Ru 5+/6+ such that Ir 5+ is oxidised before O 2− (i.e. at a lower voltage) whereas Ru 5+ is not. The voltages of the redox couples derived from dQ/dV analysis of the electrochemical load curves in Fig. 1  Electrochemistry. Self-supporting electrode films were prepared by grinding the as-synthesised materials with acetylene black and polytetrafluoroethylene in a 8:1:1 mass ratio in a pestle and mortar and subsequently calendared. Electrochemical cycling was performed in coin cells with LP30 electrolyte and a lithium metal foil counter electrode. Cells were disassembled at different states of charge and the electrodes rinsed with dry dimethylcarbonate for ex situ analysis.
Powder X-ray diffraction. Powder X-ray diffraction patterns were obtained for the as-prepared materials using a Cu source Rigaku SmartLab diffractometer equipped with a Ge(220) double bounce monochromator and without exposure to air. Reitveld profile refinements were performed using the GSAS suite of programs.
X-ray absorption spectroscopy and resonant inelastic X-ray scattering. X-ray absorption spectroscopy and resonant inelastic X-ray scattering data were obtained at the I21 beamline, Diamond Light Source. Samples were transferred to the spectrometer using a vacuum transfer suitcase to avoid air exposure and were pumped down to UHV and left to fully degas overnight. O K-edge spectra were obtained in partial fluorescence mode for bulk sensitivity. RIXS line scans were recorded at five different sample locations and averaged together. RIXS maps were collected at 0.2 eV intervals in excitation energy. All measurements were performed at 20 K to minimise any possible beam damage.  figure (b). The values of the redox couples shown in black in (b) were obtained from dQ/dV analysis extracted from the data in Fig. 1.